Explainable machine learning for predicting shellfish toxicity in the Adriatic Sea using long-term monitoring data of Harmful algae blooms

Overview

This project leverages explainable machine learning techniques to predict shellfish toxicity in the Gulf of Trieste, focusing on mussels affected by harmful algal blooms (HABs). Using a comprehensive dataset spanning 28 years, the study examines the occurrence of diarrhetic shellfish poisoning (DSP) events, enhancing the predictability and management of shellfish safety.

Objective

The goal is to develop a machine learning model that can predict DSP events accurately, helping to establish effective early warning systems for aquaculture safety.

Models Used

• Random Forest: Best overall performance for predicting DSP events.
• SVM (Support Vector Machines): Included in model evaluation for its robustness in classification tasks.
• Decision Tree Classifier: Utilized for its interpretability and ease of representation.
• MLPClassifier (Neural Networks): Explored for its capability in capturing complex patterns.

Key Findings

• Predictive Features: Certain species of phytoplankton, specifically Dinophysis fortii and D. caudata, along with environmental factors like salinity, river discharge, and precipitation, were identified as significant predictors.
• Explainability: Used methods like permutation importance and SHAP to interpret the feature influence on model predictions, ensuring transparency and understandability of the model decisions.

Dataset

This project uses a unique dataset that records the presence of toxic phytoplankton and environmental conditions over a 28-year period in the mussel farming areas of the Gulf of Trieste.

Requirements

Make sure to install the following packages:

pandas
numpy
matplotlib
seaborn
scikit-learn
shap
umap-learn
dtreeviz

Usage

1. Clone the repository.
2. Install dependencies: pip install -r requirements.txt.
3. Execute the notebooks to train models and view the analysis.

Contributing

We welcome contributions that:

• Improve prediction accuracy.
• Enhance model explainability.
• Update or expand the dataset.

License

This project is released under the MIT license.

Citation

Please cite this work in your publications if it helps your research:

@article{marzidovsek2024explainable,
  title={Explainable machine learning for predicting shellfish toxicity in the Adriatic Sea using long-term monitoring data of HABs},
  author={Marzidovšek, Martin and others},
  journal={arXiv preprint arXiv:2405.04372},
  year={2024}
}

Contact

For questions or support, please contact Martin Marzidovšek at [email protected].

Read the Full Paper

For more detailed insights and methodologies, read the full paper here: Explainable machine learning for predicting shellfish toxicity